bilateral negotiation model for agent mediated electronic commerce
DESCRIPTION
Old stuff from my master degree studyTRANSCRIPT
1
Bilateral Negotiation Model for Agent Mediated Electronic
Commerce
Gustavo Eliano de Paula
Francisco Ramos
Geber Lisboa Ramalho
Universidade Federal de Pernambuco
2
Current Bilateral Negotiation in E-commerce
• Have you got the new Guns and Roses CD?
• Yes! It costs US$15 and you can
have it in two days only!
• Could you make it for US$13 and delivering in one day?
• I can deliver within one day only if you pay US$14!
Limited to: proposals exchanges multiple attributes (e.g., price, delivery time,...)
Buyer Seller
Virtual Shop
• Ok!
3
Possible Extension: Product Amonut
• Have you got the new Guns and Roses CD?
• Yes and it costs US$15!
• Could you make it for US$13? • Only if you buy two!
Buyer Seller
Virtual Shop
Usual in real world negotiation
Hard to model.
4
Possible Extensions
Product amount
Suggestion similar product guns and roses new CD guns and Roses “Use of Illusion”
CD
Suggestion of correlated product TV & video cassette
Ultimatum this is my last offer!
Time cost selling an ice-cream in a sunny day
5
Summary
Four Problems of Modeling Bilateral Four Problems of Modeling Bilateral NegotiationNegotiation
State of the Art on Bilateral negotiationState of the Art on Bilateral negotiation
Our Negotiation ModelOur Negotiation Model
Implementation and ResultsImplementation and Results
Conclusions and Future Work.Conclusions and Future Work.
6
Four General Problems of Modeling Bilateral Negotiation
How to represent proposals? using single or multiple deal attributes?
price, taxes, ... including product attributes?
monitor size, processor speed, ...
How to evaluate proposals? attributes average
(a1 + a2 + ... + an) / n
multi-attributes utility theory (w1a1 + w2a2 + ... + wnan) / (w1+w2+...+wn)
7
Four General Problems of Modeling Bilateral Negotiation
Which are the possible moves? accept, reject (and send a counterproposal), quit give an ultimatum, suggest similar product, suggest
correlated product, etc.
How to choose the adequate move? proposal´s comparison
local decision based on current opponent proposal game theory
hard to consider complex models heuristic (Knowledge based).
8
State of the Art
KasbahKasbah Farantin’s ModelFarantin’s Model
Proposal Proposal RepresentationRepresentation
Proposal Proposal Evaluation Evaluation
Possible Possible Moves Moves
Decision Decision Making Making
Single Attribute (price)
Multiple Attributes
Price stands for proposal evaluation
Weighted attributes combination
Proposal comparison
Proposal comparison
Accept, reject, quit
Accept, reject, quit
9
State of the Art: Limitations
Kasbah and Faratin’s model proposals model is too simple moves do not consider some possibilities of a client-
salesman negotiation
Faratin’s model local deal constraints violation local deal degeneration.
10
Our Negotiation Model
Richer proposal representation
New moves (alternative product suggestion and ultimatum)
Decision making consider negotiation costs (e.g., time)
Solve Faratin’s local deals problems.
11
Proposals Representation
Proposal
Deal Attributes
Price
Delivery Time
Delivery Tax
Product Attributes
Monitor Size
Processor Speed
Fax-modem Speed
CD-ROM Speed
12
Proposals Evaluation
Deal attributes are weighted combined:
XEWPEWXPE xpcontract 21,
n
iiiiix xVwXE
1
Product attributes are weighted combined:
n
qli
qli
qlip xVwPEv
1
Deal and product evaluations are weighted combined:
For a proposal Proposalj = {Pi, Xj}, where: Pj stands for the product attributes (features) and Xj stands for the deal attributes
13
Protocol and Possible Moves
Receives
Offer
Analyzes
Offer
BuildsCounter-offer
Sends
Offer
Gives upRejectsOffer
Accepts
Offer
SuggestsAltern. Product
ChoosesAltern. Product
Builds Initial
Offer
LastOffer Sends
Last Offer
Receives
Last Offer
Analyzes
Offer
RejectsOffer
AcceptsOffer
Opponent
14
Decision making
Knowledge based consider a set of negotiation heuristics;
If oppPmyPCloseTo , oppPGoodoppPVeryGood Then oppPAccept ;
If oppPmyPCloseTo , oppPBadoppPVeryBad Then myPalNextpropos ;
If oppPmyPFarFromoppPmyPoMiddleWayT ,, Then myPalNextpropos ;
If maxtt Then myPalLastpropos, where tmax is the ,maximum time to reach a deal;If 0# mimI Then iPneSuggestioAlternativ, where # is an operator which gives the number of elements in a set;If propdl XX ,Then oppPAccept
If 0# propX Then Quit
15
Implementation and Results Implementation
100% pure Java KQML via Jatlite
A user can: negotiate by himself delegate negotiation to an autonomous agent and use
an avatar to represent him.
Facilitator
NegotiationMediator
AutonomousBuyer 1
HumanoidBuyer 1
HumanoidSeller 1
AutonomousSeller 1
avatar + figura
16
Implementation and Results
17
Implementation and Results
18
Implementation and Results
A kind of “Turing test” on different scenarios Human buyer vs. Autonomous seller Human seller vs. Autonomous buyer Human buyer vs. Human seller
Goal Check whether our negotiation model simulate human
behavior
Results Human negotiator can not identify whether its opponent
is human or autonomous.
19
Conclusions and Future Work
Contributions: extension to Faratin’s model richer proposal representation new moves (alternative product suggestion and ultimatum)
decision making considers negotiation costs solves local deals problems;
Future Features: correlated products, product quantity Learning: same product, same negotiator Funded by IBM to be integrated to WebSphere - Net
Commerce